IMPROVING DEEP LEARNING-BASED FACADE VISUAL INSPECTION: A DATA QUALITY PERSPECTIVE

Abstract

The main objective of this thesis is to improve the performance of deep learning-based façade visual inspection. In this thesis, the performance is considered from the aspects of reliability and efficiency. To achieve this objective, this thesis designs a research methodology based on the theory of total data quality management. The methodology includes four phases: definition, assessment, analysis, and improvement. The research framework is unfolded through three procedures: data selection, data annotation, and model training. For each procedure, criteria are designed to assess the data quality, and solutions are developed enabling the target stage to focus on “better” data. The experiment results demonstrate that the proposed solutions improved the accuracy and stability of the façade defects detection. Besides, the detection results obtained by the proposed solutions provide more effective outcomes for condition evaluation. Meanwhile, the time and cost are saved in general perspective because no extra labor works are expended.